Local LLM Agent Stack: A Postmortem of Real Tools for Real Workflows

This guide is your postmortem analysis of real open-source tools and workflows that local LLM users are actually using — not just benchmarks, but the actual stack that powers real agents. Whether you're running Ollama, MLX, or llama.cpp, this comparison helps you avoid tech debt and build smarter, more reliable agents. Running LLMs locally isn’t just about inference — it’s about building full workflows that include scraping, automation, and data transformation. - **Scrapy** (BSD-3-Clause): Great for large-scale crawling with robust middleware and scheduling. - **Trafilatura** (Apache-2.0/GPLv3): Lightweight, focused on extracting clean text from HTML — ideal for agent data input. - **ScrapeGraphAI** (MIT): Uses LLMs to scrape structured data directly from websites.